Using inter-colony variation in demographic parameters to assess the impact of skua predation on seabird populations


  • Conflict of interests: The authors declare no conflict of interests.

*Corresponding author.


Large skuas and gulls are top predators in marine ecosystems, feeding on shoaling fish, fishery discards and facultatively on smaller seabirds. As generalist predators they may have deleterious impacts on prey populations of seabirds, particularly when alternative foods are scarce. Declines in discards and lipid-rich shoaling fish may result in these large scavenging birds turning to prey on seabirds to meet their nutritional needs, yet we know relatively little about seabird predator–prey dynamics. Declines in Black-legged Kittiwakes Rissa tridactyla in the UK are attributed to reductions in Sandeel Ammodytes marinus availability, but may also be due to predation by Great Skuas Stercorarius skua in some parts of their range. We investigate whether variation in two demographic parameters (breeding success and population growth rate) of Kittiwake colonies across Shetland are explained by skua population density at increasing spatial scales (rings with radii of 0–5, 5–10, 10–15, 15–20, 20–25 and 25–30 km) and Kittiwake population density. These explanatory variables do not explain a significant amount of the variation in annual population growth rate (lambda), but our estimate of population change is highly conservative and we cannot exclude the possibility of type II errors. Kittiwake breeding success is positively correlated with Sandeel availability and negatively correlated with the number of Kittiwakes at the focal colony. Having controlled for these effects the number of Great Skuas also has an influence on breeding success, being negatively correlated at the scale of 5–10 and 20–25 km, but positively correlated at the scale of 10–15 km. In addition, analysis of Kittiwake populations subdivided into exposed or protected cliffs reveals that exposed sub-colonies declined more steeply than protected ones – presumably as a function of differences in susceptibility to Skua predation. We propose that comparing differences in demographic rates may be useful in unravelling seabird predator–prey dynamics, but only where there is a comprehensive demographic dataset, where it is possible to correct for confounding factors such as food availability, and information on habitat–predation interactions.

Large species of skuas and gulls are apex predators in marine ecosystems. Characterized by broad trophic niche widths, they feed primarily on shoaling fish, fishery discards and smaller seabirds. As generalist predators they may impose strong top-down regulation on prey populations of smaller seabirds, particularly when alternative food resources are unavailable (Regehr & Montevecchi 1996, Votier et al. 2004c). Many populations of these facultative predators have increased globally, and although the mechanisms involved are complex, an important contributory factor is the availability of easily accessible fishery discards (Furness 2003, Mitchell et al. 2004). However, changes in discarding policy, reductions in fishing effort and a range of technical measures will probably lead to significant declines in discard availability (Reeves & Furness 2002). Furthermore, in some areas availability of small lipid-rich shoaling fish has declined as a result of oceanographic conditions (Arnott & Ruxton 2002) or industrial fisheries (Frederiksen et al. 2004). The combination of these changes could mean a greater reliance of predatory skuas and gulls on alternative foods such as smaller seabirds (Votier et al. 2004c), but despite evidence of intense seabird predation, it remains unclear how these predators affect prey populations.

Black-legged Kittiwakes Rissa tridactyla (hereafter Kittiwake) are small, pelagic, colonial nesting seabirds, whose populations have recently shown steep declines in some parts of the UK, following marked increases during the early part of the 20th century (Mitchell et al. 2004). The causes of these declines are not clear, but changes in the availability of their main food, the Lesser Sandeel Ammodytes marinus, are likely to have played a significant role (Poloczanska et al. 2004, Frederiksen et al. 2007). Some of the most severe declines have been noted in Shetland, where the population has declined by 70% in the past two decades, compared with a 50% decline in North Sea colonies (Mitchell et al. 2004). Shetland also has the world's largest aggregation of Great Skuas Stercorarius skua, with individuals feeding facultatively on seabirds when Sandeels and fishery discards are scarce (Votier et al. 2004c). Therefore, Kittiwake declines are widely attributed to the impact of skua predation (Heubeck et al. 1997, Votier et al. 2004b). Work from one Shetland colony (Foula) revealed that Kittiwake survival was inversely related to sandeel availability and Great Skua breeding success (Oro & Furness 2002), providing support for this hypothesis. However, analysis of colony-specific variation in Kittiwake breeding success revealed a relationship with spatial differences in sandeel availability (Frederiksen et al. 2005b). Therefore, it is not clear to what extent Kittiwake population dynamics are driven by differences in food availability (‘bottom-up’ control) or by differences in skua predation (‘top-down’ control). Determining the relative strength of top-down and bottom-up processes in this predator–prey relationship is important, especially during a period when we can expect to see pronounced changes in the marine environment associated with changes in commercial fisheries and oceanographic conditions.

Here we investigate the relationship between two demographic parameters (colony growth rate and breeding success) of Kittiwake populations in Shetland and population density of their main predator, the Great Skua. Analysing inter-colony variation in demographic parameters was recently highlighted as a method to understand population dynamics better (Frederiksen et al. 2005a), although it has not, to our knowledge, been used to assess the impact of predation on seabird populations. Here we test whether Kittiwake population growth rate and breeding success are influenced by predator density at different spatial scales as well as prey density. In addition, we attempt to correct for temporal differences in food availability, but are unable to account for spatial differences. Our aim is to provide insight into the strength of top-down vs. bottom-up control of Kittiwake populations in Shetland. We also consider the value of comparing demographic rates among seabird colonies to determine impacts of predation by large species of skuas and gulls.


Data were collected from Shetland, UK, an archipelago which supports internationally important aggregations of breeding seabirds, including (in 1998–2002) approximately 7000 pairs of Great Skuas and 17 000 pairs of Kittiwakes (Mitchell et al. 2004). These two species have been studied annually in some detail as part of a long-term monitoring programme and these data form the basis for the analysis presented here.

Kittiwake demographic parameters

Population growth

All of Shetland's Kittiwakes have been counted at least once every 5 years since 1981 using standardized methods, i.e. all well-built nests capable of holding eggs or chicks are counted from a boat during peak incubation or early chick rearing in June (Heubeck et al. 1997, 1999). Colonies were defined as breeding aggregations of birds less than 1 mile apart and are listed in the Appendix. Kittiwakes breeding on Foula and Fair Isle have typically been surveyed as single colonies, when they actually consist of a number of smaller aggregations. Therefore, we excluded data from these two islands in our analysis to ensure consistency, as all other Shetland aggregations are treated separately. Not all colonies were surveyed in the same years, which could lead to bias if there are strong year effects as a consequence of annual fluctuations in food availability. To overcome this problem we used the numbers at the start and end of the study period when sampling effort was greatest to calculate the mean annual population growth rate, lambda (λ):


where N1 is the first survey (in 1981), N2 is the second (1998–2002 during a complete census of all breeding British and Irish seabirds for Seabird 2000, Mitchell et al. 2004) and t is the time interval separating them. Not all colonies were surveyed in the same year during 1998–2002, so we used only the first count in our analysis. Although this discards much intervening data it is a conservative approach that overcomes problems associated with non-random colony estimates, which could give biased estimates of changes in colony growth rate.

Breeding success

Kittiwake breeding success (number of fledged chicks per nest) has been monitored since 1986 using standard methodologies (Harris 1987) throughout Shetland, as part of the UK & Irish Seabird Monitoring Programme. We analysed data from 11 colonies for the period 1986–2005, excluding Fair Isle and Foula for the reasons given above. All but two of these colonies had data missing for certain years, which may lead to bias. To overcome problems with missing data, we used multiple imputation to provide a series of normally distributed missing values on the basis of existing data, implemented via the aregImpute function in the Hmisc package in R v.2.5.1 (R Development Core Team 2007). Existing data were fitted to a model including year and Shetland Sandeel total stock biomass (TSB) (see below) and grouped by colony.

Assessing Skua predation pressure

Data on Great Skua predation of Kittiwakes in Shetland have been collected directly via observation (Heubeck et al. 1997) and indirectly via analysis of dietary remains (Votier et al. 2004b), but only from a very small number of colonies. Therefore, we attempt to capture likely differences in the impact of Great Skua predation throughout Shetland by estimating population density at different spatial scales relative to Kittiwake colonies. We estimated the number of Great Skuas breeding within rings of radius 0–5, 5–10, 10–15, 15–20, 20–25 and 25–30 km of each focal Kittiwake colony, during 1985–87 and 1999–2002, from detailed survey data that give colony sizes at specific grid references. We predict that Kittiwake colonies closest to large Great Skua aggregations will experience higher encounter rates and therefore the greatest levels of predation. Recent research has shown that at large colonies, Great Skuas consume lower per capita amounts of seabird prey than at small colonies, probably as a function of interference competition (Votier et al. 2007). Nevertheless, the total number of seabirds consumed by Great Skuas is still higher at large colonies than at small ones.

Other factors influencing demographic parameters

Variation in demographic parameters may be influenced by a range of factors and we attempt to control for these in the hope of identifying the true influence of predation on Kittiwakes. In colonial birds the role of density dependence in the demographic process has been the subject of much debate, but is widely believed to influence population dynamics in two fundamental ways. First, at high population density piscivorous seabirds may deplete or reduce the availability of prey (Ainley et al. 2003), experience increased competition for nest-sites, or both (Coulson 1983). Secondly, at low population density colonial breeding species may be more vulnerable to predation because of poor anti-predator responses (Serrano et al. 2005, Oro et al. 2006). Therefore, maximum size of the focal Kittiwake colony was included in our analysis.

Kittiwakes in Shetland feed primarily on Lesser Sandeels (Furness & Tasker 2000) and previous research has demonstrated a strong relationship between annual variation in breeding success and Sandeel TSB (Frederiksen et al. 2007). Where appropriate (see below) we included annual Shetland Sandeel TSB estimated from a combination of virtual population analysis (VPA) during 1976–94 (ICES 2002) and research survey data (Cook 2004 and pers. comm.).

Influence of local topography on population growth rate

In addition to these large-scale explanatory variables we also investigated the influence of local topography as an indication of vulnerability to Great Skua attack. Counts from two colonies, Sumburgh Head and Troswickness, were pooled then subdivided into a total of 31 sub-colonies, each of which was classified as either being ‘exposed’ or ‘protected’. Sub-colonies within the ‘protected’ category included birds nesting in caves, very narrow geos or beneath over-hanging cliffs – areas less vulnerable to Great Skua predation than those on open cliff faces.

Statistical analysis

We investigated variation in annual population growth rate as a function of the number of Kittiwakes at the focal colony and the number of Great Skuas breeding within rings of increasing spatial scale during the 1985–87 survey, but because each colony had a single value it was not possible to include Shetland Sandeel TSB. Least-squares models were selected on the basis of Akaike's Information Criterion (AIC), which is a likelihood measure of model fit allowing for a balance between bias and precision and avoids problems of multiple testing (Burnham & Anderson 2002). We calculated ΔAIC (the difference between the preferred model, with the lowest AIC value, and the focal model) and, from this, AIC weights. AIC weights are a measure of the strength of evidence that any given model is the best descriptor of the data, with the likelihood weights summing to one. Data used in regression models were normally distributed and had equal variances.

To investigate factors influencing variation in breeding success we fitted a generalized linear mixed model (GLMM), with a quasi-Poisson error distribution (to allow for overdispersed count data with variance equal to the mean) and log link function. The number of young fledged was included as the response variable, with natural log of the sample size used as an offset and colony modelled as a random factor. We included the number of Great Skuas at increasing spatial scales as an explanatory variable in the following ways. Counts from the 1985–87 survey were fitted for the years 1986–99 inclusive and the number of Skuas from the 2000 survey fitted for the years 2000–2005 inclusive. The most recent Kittiwake population estimates for each colony and Shetland Sandeel TSB were also included as explanatory variables. Analysis was conducted by using the glmmPQL function, which is part of the MASS library in R (Venables & Ripley 2002). The penalized quasi-likelihood fitting method does not provide AIC values, and therefore models were simplified by sequentially dropping terms from the maximal model (including only two- and three-way interactions as higher order interactions do not have easily interpreted biological reasoning) until only significant terms remain (Crawley 2007).

We compared the population growth rates of Kittiwake sub-colonies exposed to risk from Great Skua attack with those protected from Great Skua attack using independent sample t-tests (assuming unequal variances). All statistical analysis was completed using R v.2.5.1 (R Development Core Team 2007).


Influence of predator and prey densities on demographic rates

Population growth rate

Comparisons among 54 colonies in Shetland revealed marked inter-colony variation in Kittiwake population growth rate. The vast majority had declined, i.e. they had a lambda of less than one (Fig. 1). The best supported model included a negative relationship between the number of Skuas within 25 km of the focal Kittiwake colony (Table 1). While this model had lower AIC values compared with the constant model, AIC weights were very similar, suggesting only a very weak influence of this explanatory variable. Other models received poorer support than the constant model, suggesting they did not explain a significant amount of the variance in annual population growth rate (Table 1).

Figure 1.

Annual population growth rates of Black-legged Kittiwake colonies in Shetland between 1981 and 1998–2002. Graph shows the frequency distribution of annual population growth rate (λ) for 54 colonies included in our analysis and illustrates both the tendency for declining populations (i.e. λ < 1) and the marked inter-colony variation within Shetland.

Table 1.  Top ten supported models selected to examine the influence of measures of Great Skua predation on annual Black-legged Kittiwake population growth rate (λ).
Dependent variable: annual population growth rate λ
ModelAICParametersΔAICAIC weight
  1. The best supported model was only slightly better supported than the null model (indicated in bold type). AIC, Akaike's Information Criterion.

1. Number of Great Skuas within 25–30 km–362.623200.180
2. Number of Great Skuas within 20–25 and 25–30 km–362.07130.5530.137
3. Constant–361.94810.6750.129
4. Number of Great Skuas within 20–25 km–361.74620.8770.116
5. Number of Great Skuas within 5–10 and 25–30 km–361.64130.9820.110
6. Number of Great Skuas within 5–10 km–361.14421.4780.086
7. Number of Great Skuas with 10–15 and 25–30 km–360.94631.6770.078
8. Number of Great Skuas within 10–15 km–360.57622.0480.065
9. Number of Great Skuas within 15–20 km–360.08622.5370.051
10. Kittiwake population–359.96222.6620.048

Breeding success

Analysis of breeding success using GLMM revealed five significant main effects in the minimal adequate model (Table 2); Sandeel TSB (positive slope =+0.000003), number of Kittiwakes at the focal colony (negative slope = –0.107), number of Great Skuas breeding within 5–10 km of the focal colony (negative slope = –0.002), number of Great Skuas breeding within 10–15 km of the focal colony (positive slope = +0.002) and number of Great Skuas breeding within 20–35 km of the focal colony (negative slope = –0.001). The slopes provide an indication of the effects of these parameters on Kittiwake breeding success. The strongest effect was the number of Kittiwakes at the focal colony. There were similar slopes for number of Great Skuas breeding within 5–10, 10–15 and 20–25 km of the focal colony but a relatively weak effect of Shetland Sandeel TSB.

Table 2.  Results of generalized linear mixed model (GLMM) (with quasi-Poisson error distribution and log link function) investigating factors influencing inter-colony variation in Black-legged Kittiwake breeding success.
  1. Number of fledged young is the dependent variable, with the number of monitored nests each year as the offset. Colony was included as a random factor, grouped within year. Bold type indicates significant terms in the minimal adequate model.

Sandeel TSB0.0000030.0000022.2490.026
Kittiwake population–0.1070.045–2.3680.019
No. of Great Skuas within 0–5 km0.00010.00030.3620.718
No. of Great Skuas within 5–10 km–0.0020.0009–2.1630.032
No. of Great Skuas within 10–15 km0.0020.00073.2910.001
No. of Great Skuas within 15–20 km–0.00090.0005–1.6080.109
No. of Great Skuas within 20–25 km–0.0010.0005–2.1050.037
No. of Great Skuas within 25–30 km0.00030.00021.7030.090

Influence of local topography on population growth rate

All sub-colonies at Sumburgh Head and Troswickness declined during the period 1990–2006. However, declines were significantly influenced by local topography (t-test assuming unequal variances: t = –3.1378, df = 6.29, P = 0.0189), with exposed sub-colonies declining significantly more steeply (mean λ = 0.7146, sd = ± 0.1613) compared with protected colonies (mean λ = 0.9082 ± 0.0462) (Fig. 2).

Figure 2.

Change in population of Black-legged Kittiwake breeding aggregations subdivided by topography at Troswickness and Sumburgh Head, Shetland, 1990–2006. For the two colonies, nesting aggregations are clumped by whether they occur on (a) exposed cliffs or in (b) protected areas (in caves or beneath overhangs) and therefore provide a measure of their vulnerability to skua attack.


Analysis of Kittiwake population growth rates in Shetland revealed strong inter-colony variation (Fig. 1), but this variation was poorly explained by the number of and proximity to Great Skuas as well as the number of Kittiwakes breeding at the focal colony (Table 1). The best supported model included a negative effect of the number of Great Skuas breeding within 25–30 km of the focal colony (r = 0.219, n = 53, P = 0.110). However, this model is almost indistinguishable from the null model, so there is no convincing evidence here that population growth rates in Kittiwake colonies throughout Shetland are affected by Skua predation, despite its assumed negative impact (Heubeck et al. 1997, 1999). Instead, differences in annual population growth rate may be more strongly influenced by regional variation in the availability of their primary food, Lesser Sandeels (Frederiksen et al. 2005b), or alternative local food sources such as fishery discards (Pennington et al. 2004). Indeed, an obvious weakness in our analysis was the inability to control for temporal variation in Sandeel abundance both temporally and spatially. Although we attempted to model growth rate controlling for annual fluctuations in Sandeel abundance, we were unable to overcome problems of temporal autocorrelation satisfactorily while fitting the appropriate model structure. Further work using more sophisticated modelling approaches to incorporate Sandeel abundance might be fruitful, as would the inclusion of spatial variation in Sandeel availability at the colony level. In addition, our measure of Great Skua predation pressure may not accurately reflect their impact on Kittiwake populations. Analysis of dietary remains reveals that Great Skuas show distinct dietary differences both among (Phillips et al. 1997) and within (Votier et al. 2004a) colonies, which may be a function of spatial variation in food availability or different learned foraging behaviour. In addition, there is evidence of heavy seabird predation in areas distant from Great Skua colonies (Heubeck et al. 1997), which may represent non-breeding birds or breeders travelling considerable distances to forage on seabird colonies. In either instance our measures of Great Skua predation would not be likely to capture these specific differences in predatory behaviour, thus reducing our ability to detect top-down effects. Kittiwake breeding success was positively correlated with the Shetland Sandeel TSB, and negatively correlated with the number of Kittiwakes at the focal colony (Table 2). These results are consistent with previous studies linking breeding success with both Sandeel abundance (e.g. Poloczanska et al. 2004) and number of conspecifics (Coulson 1983, Ainley et al. 2003), revealing the importance of food availability in determining reproductive success. There was a negative correlation between breeding success and the number of Great Skuas breeding within both a 5–10-km and a 20–25-km radius of the focal colony (Table 2). Reduced success may be a function of the impact of Great Skua predation or competition between Kittiwakes and Great Skuas for the same food (Sandeels and fishery discards). However, observations indicate that Great Skuas regularly kill both young and adult Kittiwakes and therefore it seems likely that Great Skua predation does have a deleterious impact. Unexpectedly, the number of Great Skuas immediately adjacent to Kittiwake colonies (i.e. in the 0–5-km ring) did not emerge as a significant term in our analysis. A positive relationship between Kittiwake breeding success and the number of Great Skuas breeding within 10–15 km is also difficult to explain, but may be related to spatial differences in food availability. In summary, Kittiwake breeding success was influenced by food availability (directly or indirectly via intra-specific competition) and there is some evidence that proximity to Great Skuas may detrimentally affect Kittiwake breeding success, as predicted. However, given the significant correlation between Kittiwake breeding success and number of Great Skuas it is difficult to draw firm conclusions from these results.

Comparing population growth rates at two colonies subdivided into exposed areas and areas protected from Great Skuas by caves or overhangs, revealed that, although all areas showed declines, nesting aggregations in protected areas declined significantly less steeply than in exposed areas (Fig. 2). Although exposed areas of cliffs may be more prone to deleterious weather effects compared with areas in caves or beneath overhangs, extensive anecdotal observations suggest that Great Skuas are efficient predators of adult and fledgling Kittiwakes on exposed cliff faces, but have difficulty foraging within enclosed sections of cliff (M. Heubeck and S.C. Votier pers. obs.). It is also possible that other avian predators (primarily Herring Gulls Larus argentatus, Ravens Corvus corax and Hooded Crows C. cornix) may play a role in explaining these patterns. However, predation by these species was infrequent and tended to be highly localized (e.g. Heubeck & Mellor 1994). We believe that the observed differences are most likely the result of differential Great Skua predation and provide further evidence of top-down control on Kittiwake colonies. Although Kittiwakes generally have a high level of site fidelity once established at a breeding site (Danchin et al. 1998), birds are more likely to change colonies following predation and disturbance events (Coulson & Nève de Mévergnies 1992, Danchin & Monnat 1992). Therefore, declines at exposed sections of cliff may be more the result of dispersal to avoid disturbance by predators, rather than a direct result of mortality of adult Kittiwakes. Periodic increases at some protected sites support this (Fig. 2b; Heubeck & Mellor 1997).

Using inter-colony variation in demographic rates to assess predation impact

Our results provide some evidence that Kittiwake breeding success is negatively affected by Great Skua density at a relatively local (5–10 km) scale. However, previous analysis of breeding success (not shown), which did not include imputed missing values, failed to detect similar patterns, indicating the importance of a complete or near complete dataset. In addition, there was a significant positive effect of Great Skua density at 10–15 km on Kittiwake breeding success. Therefore, the impact of Great Skua predation on Kittiwake breeding success is unclear. Based on the slope of two colony counts to estimate annual population growth rate, we found no strong links with Kittiwake or Great Skua numbers. However, this may have been masked by temporal and spatial variation in food availability and much intervening data have been lost using this conservative approach to analysing population trends. Although software does exist to fit unbiased population trends which impute missing values and allow for autocorrelation (some GLMM and freeware TRIM), we were unable to fit our data to these models. In addition, differences in colony growth rate between exposed and protected areas of cliff were significant, indicating an influence of Great Skua predation. The impact of topography should therefore be borne in mind when determining predator–prey dynamics of cliff-nesting seabirds. Using variation in demographic parameters to assess the impact of predation at other locations may be revealing but interpretation of results should be viewed cautiously. In Shetland much has changed in terms of the numbers of Great Skuas and Kittiwakes, as well as their food, which is reflected in a highly fluid system. Future work would greatly benefit from quantifying availability of marine prey on the scale of tens of km2 together with a more detailed understanding of interactions between topography and Great Skua foraging behaviours. Together with these data and more detailed data on prey population demographics it may be possible to quantify more accurately the complex interaction between avian predator, prey and food availability.

We thank all those individuals involved with collecting data as part of the Joint Nature Conservation Committee (JNCC) partnership with the Shetland Oil Terminal Environmental Advisory Group (SOTEAG) and the Royal Society for the Protection of Birds, as well as Scottish Natural Heritage, and Mick Mellor (SOTEAG). We particularly thank Norman Ratcliffe and Rob Robinson who spent much time helping with the analysis, as well as Roddy Mavor and Matt Parsons at JNCC for providing Kittiwake breeding success data. Robin Cook (FRS Marine Laboratory) kindly provided help with Shetland Sandeel data. An anonymous referee and Pete Cotton provided useful comments on the manuscript and Kirsten Archibald assisted with data analysis.

Table Appendix..  List of the 60 Black-legged Kittiwake breeding aggregations surveyed in Shetland, UK, 1981 and again during 1998–2002.
Sumburgh HeadGloup Holm
Horse IslandBlue Mull
Siggar NessLang Holm
Fitful HeadSouth Holms
St Ninian's IsleHermaness
Ness of IrelandMuckle Flugga
South HavraSaxavord
Kettla NessVirdik
West BurraBurgar
Reawick NessMooa Stack
Skelda NessRamaberg
WesterwickClett Stack
Burga StacksStrandburgh Ness
VailaSouth–east Fetlar
Braga NessLambhoga
Wats NessErne Stack/Vatsetter
Papa StourBurravoe
Turl StackGrunay
Muckle Roe SouthNorth Benelip
Muckle Roe NorthClett Head
Dore HolmLevaneap
Skerry of EshanessNoss
Eshaness CliffsMillburn Geo, Bressay
TingonHole of Bugars
Gruna StackMousa
FethalandTroswick Ness
Ramna StacksBoddam Caves
VarnadilFair Isle